楊偉松,楊德軍
?
二維網(wǎng)格上經(jīng)紀(jì)人模仿引起的自組織分離與聚合
楊偉松1,楊德軍2
(1. 江西科技師范大學(xué) 通信與電子學(xué)院,江西 南昌 330013;2. 湖北文理學(xué)院 教育學(xué)院,湖北 襄陽(yáng) 441053)
將“演化的爭(zhēng)當(dāng)少數(shù)者博弈”模型(EMG)建立在41×41的二維正方形網(wǎng)格上,格點(diǎn)代表經(jīng)紀(jì)人,連線代表該兩相鄰經(jīng)紀(jì)人存在聯(lián)系,每個(gè)經(jīng)紀(jì)人被隨機(jī)分配一個(gè)介于0與1之間的基因策略值。數(shù)值模擬結(jié)果表明:模型的獎(jiǎng)懲比等于或大于1時(shí),隨著演化過程的進(jìn)行,系統(tǒng)經(jīng)紀(jì)人傾向于模仿具有極端基因策略值(接近0或1)經(jīng)紀(jì)人的策略,說明極端的決定策略比猶豫策略(接近0.5)的表現(xiàn)要好,能使經(jīng)紀(jì)人獲得更大收益;當(dāng)經(jīng)紀(jì)人策略自組織分離現(xiàn)象出現(xiàn)后,人群-反人群效應(yīng)更加明顯,系統(tǒng)一方人數(shù)的變化偏差顯著降低,系統(tǒng)資源得到更加有效的利用. 而當(dāng)模型的獎(jiǎng)懲比小于1時(shí),即處于困難時(shí)期時(shí),中間猶豫策略的表現(xiàn)則比極端策略要好.
EMG模型;二維網(wǎng)格;經(jīng)紀(jì)人模仿;人群-反人群效應(yīng);自組織分離與聚合
從MG及EMG模型提出以來,有許多研究者在其基礎(chǔ)上做了些引入模仿的工作. 如Slanina研究了一維周期鏈結(jié)構(gòu)中的模仿少數(shù)者博弈模型[1-2],Quan等人研究了模仿合金少數(shù)者博弈模型[3]及雙向模仿的演化少數(shù)者博弈模型[4]. 他們發(fā)現(xiàn)經(jīng)紀(jì)人分布在一維周期鏈上通過相鄰經(jīng)紀(jì)人的策略模仿會(huì)提高模型系統(tǒng)的整體效率.
本文假設(shè)將EMG模型的經(jīng)紀(jì)人放在二維網(wǎng)格上,考察通過相鄰經(jīng)紀(jì)人之間的模仿以及在不同的環(huán)境獎(jiǎng)懲比條件下系統(tǒng)中經(jīng)紀(jì)人的策略分布變化狀況以及系統(tǒng)整體效率的演化情況.
在D. Challet等人提出“爭(zhēng)當(dāng)少數(shù)者博弈”模型(MG叫做)[5-6]的基礎(chǔ)上,N. F. Johnson等人提出了一種“演化的爭(zhēng)當(dāng)少數(shù)者博弈”模型(EMG)[7]. 模型如下:
本文假設(shè)經(jīng)紀(jì)人分布在一個(gè)41×41的二維正方形網(wǎng)格上,每個(gè)格點(diǎn)代表一個(gè)經(jīng)紀(jì)人,連線代表該兩相鄰經(jīng)紀(jì)人存在聯(lián)系. 這樣除了邊界線上的格點(diǎn)之外,每個(gè)格點(diǎn)一共有上下左右四個(gè)鄰居.
根據(jù)模型設(shè)定規(guī)則,用FORTRAN語言編程計(jì)算策略分別在從0到1的10個(gè)小區(qū)間(區(qū)間長(zhǎng)度為0.1)內(nèi)的經(jīng)紀(jì)人數(shù)目,并用Origin軟件畫圖. 開始時(shí)系統(tǒng)經(jīng)紀(jì)人的策略分布情況見圖1,經(jīng)過一段時(shí)期(10000000時(shí)步)的演化后,系統(tǒng)中經(jīng)紀(jì)人的基因策略值出現(xiàn)自組織分離現(xiàn)象,結(jié)果見圖2.
圖1 開始時(shí)系統(tǒng)中經(jīng)紀(jì)人的基因策略分布
圖2 經(jīng)過10000000時(shí)步演化后系統(tǒng)中經(jīng)紀(jì)人的基因策略分布
從圖1和圖2的結(jié)果來看,經(jīng)過10000000時(shí)間的演化后,經(jīng)紀(jì)人傾向于擁有兩個(gè)極端策略中的一個(gè),在策略自組織分離現(xiàn)象出現(xiàn)后,會(huì)形成典型的人群-反人群效應(yīng),從而有效降低系統(tǒng)某方人數(shù)的變化方差,這意味著少數(shù)方或獲勝方的人數(shù)會(huì)增加,更多的人會(huì)享受到利益,從而提高系統(tǒng)整體資源的利用效率.
此外,以1000時(shí)步為一代,讓系統(tǒng)運(yùn)行10000代,并統(tǒng)計(jì)了每一代中系統(tǒng)一方人數(shù)的波動(dòng)標(biāo)準(zhǔn)偏差,標(biāo)準(zhǔn)偏差的定義如下:
系統(tǒng)標(biāo)準(zhǔn)偏差隨演化代數(shù)的變化關(guān)系見圖3.
圖3 系統(tǒng)一方人數(shù)的波動(dòng)標(biāo)準(zhǔn)偏差隨演化代數(shù)的變化關(guān)系
當(dāng)假設(shè)模型系統(tǒng)的獎(jiǎng)懲比小于1時(shí),即設(shè)經(jīng)紀(jì)人進(jìn)入少數(shù)方獲得的獎(jiǎng)勵(lì)為0.8分,進(jìn)入多數(shù)方得到的懲罰為1分時(shí),數(shù)值模擬計(jì)算發(fā)現(xiàn),經(jīng)過一段時(shí)期(10000000時(shí)步)的模仿演化后,系統(tǒng)中經(jīng)紀(jì)人的基因策略值會(huì)出現(xiàn)與自組織分離相反的現(xiàn)象,即趨向集中于采用中間策略.
進(jìn)一步考慮獎(jiǎng)懲比情況更強(qiáng)烈的情況:如獎(jiǎng)懲比更大如1.2時(shí),模型系統(tǒng)經(jīng)過長(zhǎng)時(shí)期(10000000時(shí)步)演化后,經(jīng)紀(jì)人策略分布狀況與獎(jiǎng)懲比等于1的模型系統(tǒng)的情況類似;當(dāng)獎(jiǎng)懲比更小如0.6時(shí),模型系統(tǒng)經(jīng)過長(zhǎng)時(shí)期(10000000時(shí)步)演化后,經(jīng)紀(jì)人策略分布狀況則與獎(jiǎng)懲比等于0.8的模型系統(tǒng)的情況類似. 說明安樂時(shí)期(獎(jiǎng)懲比大于等于1)與困難時(shí)期(獎(jiǎng)懲比小于1)系統(tǒng)經(jīng)紀(jì)人的優(yōu)勢(shì)策略存在本質(zhì)的區(qū)別.
圖4 獎(jiǎng)懲比為0.8時(shí)經(jīng)過10000000時(shí)步演化后系統(tǒng)中經(jīng)紀(jì)人的基因策略分布
圖5 獎(jiǎng)懲比分別為1.2和0.6時(shí)經(jīng)過10000000時(shí)步演化后系統(tǒng)中系統(tǒng)中經(jīng)紀(jì)人的基因策略分布
文章數(shù)值模擬的時(shí)步比傳統(tǒng)模型要長(zhǎng),結(jié)果也更可靠. 從數(shù)值模擬結(jié)果來看,當(dāng)模型的獎(jiǎng)懲比等于或大于1時(shí),隨著演化過程的進(jìn)行,系統(tǒng)經(jīng)紀(jì)人傾向于模仿具有極端基因策略值(接近0或1)經(jīng)紀(jì)人的策略,說明極端的決定策略比猶豫策略(接近0.5)的表現(xiàn)要好,能使經(jīng)紀(jì)人獲得更大收益. 當(dāng)經(jīng)紀(jì)人策略自組織分離現(xiàn)象出現(xiàn)后,人群-反人群效應(yīng)更加明顯,系統(tǒng)一方人數(shù)的變化偏差顯著降低,系統(tǒng)資源得到更加有效的利用. 而當(dāng)模型的獎(jiǎng)懲比小于1時(shí),即處于困難時(shí)期時(shí),中間猶豫策略的表現(xiàn)則比極端策略要好.
[1] SLANINA FRANTI?EK. Social organization in the minority game model[J]. Physica A: Statistical Mechanics and its Applications, 2000, 286(1-2): 367-376.
[2] SLANINA FRANTI?EK. Harms and benefits from social imitation [J]. Physica A: Statistical Mechanics and its Applications, 2001, 299(8-10): 334-343.
[3] QUAN HONGJUN, WANG BINGHONG, HUI PAK-MING, et al. Cooperation in the mixed population minority game with imitaiton[J]. Chinese Phys. Lett., 2001, 18(9):1156-1158.
[4] QUAN HONGJUN, WANG BINGHONG, HUI PAK-MING. Effects of imitation in a competing and evolving population[J]. Physica A: Statistical Mechanics and its Applications, 2002, 312(3-4): 619-626.
[5] CHALLET D, ZHANG Y C. Emergence of cooperation and organization in an evolutionary game[J]. Physica A: Statistical Mechanics and its Applications, 1997, 246(3-4): 407-418.
[6] CHALLET D, ZHANG Y C. On the minority game: Analytical and numerical studies[J]. Physica A: Statistical Mechanics and its Applications, 1998, 256: 514-532.
[7] JOHNSON NEIL F, HUI PAK MING, JONSON ROB, et al. Self-organized segregation within an evolving population[J]. Phys. Rev. Lett., 1999, 82: 3360-3363.
Self-segregation and Clustering Induced by Agents’ Imitation on Two-dimensional Lattice
YANG Weisong1, YANG Dejun2
(1. College of Communication and Electronics, Jiangxi Science & Technology Normal University, Nanchang 330013, China; 2.College of Education, Hubei University of Arts and Science, Xiangyang 441053, China )
With EMG model on 41×41 two-dimensional lattice, lattice point represents agent, ligature represents communication of two neighbouring agents, and each agent is assigned a gene strategy p value between 0 and 1 randomly. Numerical simulation shows that when prize-fine ratio of model is equal to or larger than 1, with proceeding of evolution, agents in system tend to imitate the strategy of agents who have extreme gene strategy(near 0 or 1), which shows extreme strategy performs better than hesitate strategy(near 0.5), and may brings more benefit to agents; when self-segregation phenomenon of agent strategy appears, crowd-anticrowd effect is more evident, the variance of the number of agents in one side reduces significantly, the system resource is utilized more efficiently. When the prize-fine ratio is less than 1, namely in difficult period, middle hesitate strategy performs better than the extreme strategy.
EMG model; Two-dimensional lattice; Agent imitation; Crowd-anticrowd effect; Self-segregation and clustering
O225
A
2095-4476(2014)05-0021-03
2013-12-05;
2014-03-20
楊偉松(1977— ), 男, 江西南昌人, 江西科技師范大學(xué)通信與電子學(xué)院講師.
(責(zé)任編輯:饒 超)